Trading off subjective values and social expectations for interpersonal trust

Trust is central to a large variety of social interactions. Different research fields have empirically and theoretically investigated trust, observing trusting behaviors in different situations and pinpointing its different components and constituents. However, a unifying, computational formalization of those diverse components and constituents of trust is still lacking. Previous work has mainly used computational models borrowed from other fields and developed for other purposes to explain trusting behaviors in empirical paradigms. In particular, most of those models propose that social behaviors are generated by subjective utility functions that trade off the expected benefits and costs of the prosocial act. However, these models overlook knowledge from social psychology suggesting that individuals weigh different social expectations (e.g., that others have of them and they themselves have of others) before engaging in a variety of social behaviors. Here, I propose that a computational model for social behaviors should combine current and prospective action values with social beliefs and expectancies about one’s and a partner’s behavior. Hence, the proposed model extends previous cost/utility functions by accounting for social expectations about a partner’s likely, future behavior. I then showcase how this can be done by computationally formalizing a verbal model of trust, that is, the vulnerability model that defines trust as the willingness to accept vulnerability to others based on their trustworthiness. In the context of the classic investment game (IG)––an economic game thought to capture some important features of trusting behaviors in social interactions (Figure 1), I show how variations of a single parameter of the vulnerability model that governs the balance between a scalar subjective value and learnt social expectations, generates varying behaviors that can be interpreted as stemming from different “trust attitudes”. I then show how these behavioral patterns change as a function of an individual’s loss aversion and expectations of a partner’s behavior, allowing for changes in trust contingent on the partner’s trustworthiness like psychological models predict. I finally demonstrate how the vulnerability model can be easily extended in a novel IG paradigm to investigate different expectations about an interacting partner based on different types of information that contribute to expectations of the partner’s trustworthiness. In particular, I will focus on benevolence and competence (Figure 2)––two character traits central to trustworthiness impressions. I will show how different trusting behaviors will be generated by different weights on a partner’s benevolence and competence, and how the model predicts that people should prefer a benevolent but incompetent partner over a malevolent one. The vulnerability model can be employed as is or as a utility function within more complex Bayesian frameworks for inferences in different social environments where actions are associated with subjective values and weighted by individual expectations of others’ behaviors. This work provides an important building block for future theoretical and empirical work across a variety of research fields